Human-on-the-loop (HOTL) lets an AI system act while a person monitors it in real time and can intervene. Unlike human-in-the-loop (HITL), it does not require approval before each action. section 4.7.1
Use HOTL alongside boundaries that limit what the system can do and how much damage it can cause:
- Exclude sensitive operations the AI must never perform, such as changing contract terms or deleting records without recovery options. section 4.7.1
- Set rate and scope limits, including caps on actions, affected records, transaction value, or resources modified. section 4.7.1
- For high-impact actions, require users to verify each critical parameter before proceeding. section 4.7.1
